Auto-associative extreme learning factory as a single class classifier
Vadlamani Ravi, Puneet Singh · 2014
We first design the Auto Associative Extreme Learning Machine (AAELM) as an auto associative version of the ELM and then propose a single class classifier based on Auto Associative Extreme Learning Factory (AAELF), which is an ensemble of several AAELMs. The ensemble was necessitated because the results of AAELM are extremely sensitive to the random weights of the connections between input and hidden layers. The proposed architecture is tested on bankruptcy prediction datasets namely Spanish banks, Turkish banks, UK banks; UK Credit dataset and phishing dataset. It turns out that AAELF outperforms past works that included many binary and single class classifiers. It is concluded that AAELF is an effective single class classifier in classifying highly unbalanced datasets or datasets where the positive class is totally missing.